Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method of predicting a plant control output by an adaptive computerized predictor apparatus, the method comprising: configuring the adaptive computerized predictor apparatus, using one or more processors, to operate in accordance with a learning process based on a teaching input; at a first time instance, based on a sensory context, causing the adaptive computerized predictor apparatus to generate a first predicted plant control output; configuring the adaptive computerized predictor apparatus, using the one or more processors, to provide the first predicted plant control output as the teaching input into the learning process; at a second time instance subsequent to the first time instance, causing the adaptive computerized predictor apparatus to generate a second predicted plant control output based on the sensory context and the teaching input; adjusting the learning process based on a difference between the second predicted plant control output and the teaching input; and causing a plant to perform an action consistent with the sensory context and the adjusted learning process, the action being in accordance with a defined target trajectory; wherein the teaching input comprises the first predicted plant control output.
A computerized system predicts plant control output by learning from its own actions. The system initially generates a plant control output based on sensory input. This output is then fed back into the system as a "teaching input" to refine the learning process. Subsequently, the system generates a second output based on both the same sensory input and the teaching input. The learning process is adjusted based on the difference between this second output and the original teaching input. The "plant" then performs an action based on the sensory input and the adjusted learning, attempting to follow a defined trajectory.
2. The method of claim 1 , wherein: the plant comprises a robotic platform; responsive to the sensory context comprising a representation of an obstacle, the action comprises executing an avoidance maneuver by the robotic platform; and responsive to the sensory context comprising a representation of an target, the action comprises executing an approach maneuver by the robotic platform.
The system from the previous description controls a robotic platform. When the sensory input represents an obstacle, the action taken by the robot is an avoidance maneuver. Conversely, when the sensory input represents a target, the robot executes an approach maneuver. The system thus learns to control the robot's movements in response to different sensory inputs representing the environment, autonomously navigating towards targets and away from obstacles by adapting its control signals based on the feedback loop.
3. The method of claim 1 , further comprising basing the sensory context on sensory input into the learning process, a portion of the sensory input comprising a video sensor data and another portion of the sensory input comprising the predicted plant control output.
The sensory context, which drives the control output, is based on sensory input fed into the learning process. Part of this input comes from video sensor data, giving the system visual information about its surroundings. Another part of the sensory input is the plant control output itself, providing the system with information about its own actions. This combination allows the system to learn relationships between what it "sees," what it "does," and the consequences of those actions.
4. The method of claim 1 , wherein the learning process comprises adapting a network of computerized neurons in accordance with the sensory context and the teaching input.
The learning process involves adapting a network of computerized neurons. The connections and behavior of these neurons are adjusted based on both the sensory input (representing the environment) and the teaching input (representing the system's own past actions). This allows the network to learn complex relationships and generate appropriate plant control outputs for different scenarios.
5. The method of claim 4 , further comprising: interconnecting multiple ones of the network of computerized neurons with connections each characterized by a connection efficacy; and the adapting the network of computerized neurons comprises adapting the connection efficacy of individual connections based on the sensory context and the teaching input.
The computerized neuron network from the previous description comprises multiple interconnected neurons, each connection having a "connection efficacy" representing its strength. The system adjusts these efficacies during learning based on the sensory context and the teaching input. Stronger connections, and therefore more influence of one neuron on another, are learned for scenarios which give correct control signal.
6. The method of claim 4 , wherein the adapting the network of computerized neurons is based on an error measure between the predicted plant control output and the teaching input.
The computerized neuron network's adaptation is based on an error measurement. This measurement quantifies the difference between the predicted plant control output (the system's action) and the teaching input (the desired or previous action). The network adjusts its connections and parameters to minimize this error, driving the system towards more accurate and effective control.
7. The method of claim 4 , further comprising: communicatively coupling individual ones of the network of computerized neurons to connections characterized by a connection efficacy; wherein individual ones of the network of computerized neurons are configured to be operable in accordance with a dynamic process characterized by an excitability parameter; basing the sensory context on input spikes delivered to the adaptive computerized predictor apparatus via a portion of the connections, individual ones of the input spikes being capable of increasing the excitability parameter associated with individual ones of the network of computerized neurons; and wherein the teaching input comprises one or more teaching spikes configured to adjust an efficacy of the portion of the connections, an efficacy adjustment for a given connection providing a portion of the input spikes to a given computerized neuron being configured based on one or more events occurring within a plasticity window, the one or more events including one or more of: (i) a presence of one or more input spikes on the given connection, (ii) an output being generated by the given computerized neuron, or (iii) an occurrence of at least one of the one or more teaching spikes.
Individual computerized neurons in the learning network are coupled via connections characterized by a connection efficacy. Each neuron has an "excitability parameter" determining its output. Sensory input is delivered as "input spikes" increasing neuron excitability. The "teaching input" takes the form of "teaching spikes" that adjust the efficacy of connections. Efficacy adjustment is based on events within a "plasticity window": presence of input spikes, neuron output, or teaching spikes. These events dictate how the connection strength is modified, enabling spike-timing-dependent plasticity.
8. The method of claim 7 , wherein, responsive to the sensory context being updated at 40 ms intervals, selecting a plasticity window duration from a range between 5 ms and 200 ms, inclusive.
In the spike-based computerized neuron network, with sensory input updated every 40ms, the "plasticity window" duration for adjusting connection strengths is chosen from a range of 5ms to 200ms. This window determines the timeframe within which events like input spikes, neuron firing, and teaching spikes influence connection efficacy, thereby tuning the network's learning dynamics to the input update rate.
9. The method of claim 4 , wherein: a portion of the network of computerized neurons comprise spiking neurons, individual ones of the spiking neurons being characterized by a neuron excitability parameter configured to determine an output spike generation by a corresponding spiking neuron; multiple ones of the spiking neurons is interconnected by connections characterized by second connection efficacy, individual ones of the connections being configured to communicate one or more spikes from one or more pre-synaptic spiking neurons to one or more post-synaptic spiking neurons; and a portion of the sensory context is based on sensory input into the learning process comprising the one or more spikes.
The computerized neuron network uses "spiking neurons" with an "excitability parameter" controlling output spike generation. These neurons are interconnected with connections characterized by "connection efficacy," transmitting spikes between pre- and post-synaptic neurons. Sensory input to the learning process is also represented as these spikes. This enables the system to process and learn from temporal patterns in the sensory data.
10. The method of claim 9 , wherein the causing the adaptive computerized predictor apparatus to generate the first or the second predicted plant control output comprises generating one or more other spikes based on spike outputs by individual ones of the spiking neurons.
Generating the predicted plant control output (first or second) in the spiking neural network system involves generating one or more other spikes based on spike outputs from the individual spiking neurons within the network. The system converts the distributed spike activity of the network into a specific control signal for the plant, effectively translating neural activity into actionable commands.
11. The method of claim 9 , further comprising communicating the sensory input via a portion of the connections via one or more other spikes.
Sensory input in the spiking neural network system is communicated via a portion of the connections using one or more other spikes. This means that the sensory information is encoded and transmitted through the network in the form of spike trains, allowing the system to process and integrate sensory data directly within its spiking neuron architecture.
12. The method of claim 9 , wherein: the predicted plant control output comprises a continuous signal configured based on one or more spike outputs by the individual ones of the spiking neurons; and the continuous signal includes one or more of an analog signal, a polyadic signal with arity greater than one, an n-bit long discrete signal with n-bits greater than one, a real-valued signal, or a digital representation of a real-valued signal.
The predicted plant control output is a continuous signal (analog, polyadic, n-bit discrete, or real-valued) configured based on spike outputs from the individual spiking neurons. The spiking neuron network outputs discrete events, but the final control signal is converted to a continuous form usable by the "plant," like a robot motor.
13. The method of claim 9 , wherein: the sensory input comprises a continuous signal; and the continuous signal includes one or more of an analog signal, a polyadic signal with arity greater than 1, an n-bit long discrete signal with n-bits greater than 1, or a real-valued signal, or a digital representation of an analog signal.
The sensory input is provided as a continuous signal (analog, polyadic, n-bit discrete, or real-valued). Sensory information that may come from physical measurement is converted into a format usable by the system.
14. The method of claim 9 , wherein the sensory input comprises a binary signal characterized by a single bit.
The sensory input is a binary signal characterized by a single bit. This means the system can operate with extremely simple input, potentially representing the presence or absence of a feature, which it uses to drive its learning and control process.
15. The method of claim 1 , further comprising: updating the learning process at regular time intervals; and adapting a network of computerized neurons based on an error measure between (i) the predicted plant control output generated at a given time instance and (ii) the teaching signal determined at another given time instance prior to the given time instance, the given time instance and the another time instance being separated by a duration equal to one of the regular time intervals.
The learning process is updated at regular time intervals. The system adapts its network of computerized neurons based on an error measure, comparing the plant control output generated at a given time to the teaching signal from an earlier time, separated by one of these regular time intervals. This introduces a time delay into the learning process.
16. The method of claim 1 , wherein: the plant comprises at least one motor comprising a motor interface; and the predicted plant control output comprises one or more instructions to the motor interface configured to actuate the at least one motor.
The "plant" being controlled includes at least one motor with a motor interface. The predicted plant control output consists of instructions sent to this interface, designed to actuate the motor. Therefore, the system's output is directly controlling the movement or operation of a physical device via motor commands.
17. The method of claim 1 , wherein the learning process comprises a supervised learning process.
The learning process is a supervised learning process. This implies that the system is trained with labeled data, where the correct or desired plant control output is provided alongside the sensory input, allowing the system to learn the mapping between the two.
18. The method of claim 1 , wherein the predicted plant control output comprises a vector of outputs comprising two or more output components.
The predicted plant control output is a vector of outputs, comprising two or more output components. This indicates the system can control multiple aspects of the "plant" simultaneously, by generating a multi-dimensional control signal, such as controlling multiple joints of a robot arm.
19. The method of claim 1 , wherein the learning process is configured based on one or more of a look up table, a hash-table, a data base table configured to store a relationship between a given sensory context, a given teaching input associated with the given sensory context, and the predicted plant control output generated for the given sensory context during learning.
The learning process is configured using a lookup table, hash table, or database table. This table stores the relationship between a given sensory context, its associated teaching input, and the predicted plant control output generated during learning. The system retrieves appropriate outputs based on sensory context.
20. A non-transitory computer-readable medium comprising instructions stored thereon, the instructions being configured to, when executed by a processing apparatus, cause the processing apparatus to: initialize a learning process based on a teaching input; generate a first predicted plant control output at a first time instance, based on a sensory context; provide the first predicted plant control output as the teaching input into the learning process; generate a second predicted plant control output based on the sensory context and the teaching input at a second time instance subsequent to the first time instance; and adjust the learning process based on an error measure between the predicted plant control output and the teaching input; wherein the predicted plant control output is configured to cause the plant to perform an action consistent with the sensory context and the learning process; wherein the learning process is configured based on a network of computerized neurons configured to be adapted in accordance with the sensory context and the teaching input; and wherein the adaptation of the network of computerized neurons is based on the error measure between the predicted plant control output and the teaching input.
A computer-readable medium stores instructions for predicting plant control output. The instructions initialize a learning process based on a teaching input, generate a first predicted plant control output based on sensory context, provide the first output as a teaching input, generate a second predicted output, and adjust the learning based on the error between the predicted output and teaching input. The plant performs actions based on the context and learning process. The learning is based on a computerized neuron network adapted based on context and teaching input, with adaptation driven by the error between predicted output and teaching input.
21. The non-transitory computer-readable medium of claim 20 , wherein the instructions are further configured to, when executed, cause the processing apparatus to: receive the sensory context via one or more sensors; and provide the teaching input via a controller.
The non-transitory computer-readable medium instructions from the previous description further receive the sensory context from one or more sensors and the teaching input is provided by a controller. This specifies that the sensory input is actively acquired from an external environment through sensors and the teaching signals from an external source, highlighting the interplay of hardware and software components.
Unknown
September 19, 2017
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.